sys/kern/imgact_elf.c

   1 /*-
   2  * Copyright (c) 2000 David O'Brien
   3  * Copyright (c) 1995-1996 Søren Schmidt
   4  * Copyright (c) 1996 Peter Wemm
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer
  12  *    in this position and unchanged.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. The name of the author may not be used to endorse or promote products
  17  *    derived from this software without specific prior written permission
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29  */
  30
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include "opt_compat.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/exec.h>
  38 #include <sys/fcntl.h>
  39 #include <sys/imgact.h>
  40 #include <sys/imgact_elf.h>
  41 #include <sys/kernel.h>
  42 #include <sys/lock.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mount.h>
  45 #include <sys/mutex.h>
  46 #include <sys/mman.h>
  47 #include <sys/namei.h>
  48 #include <sys/pioctl.h>
  49 #include <sys/proc.h>
  50 #include <sys/procfs.h>
  51 #include <sys/resourcevar.h>
  52 #include <sys/sf_buf.h>
  53 #include <sys/systm.h>
  54 #include <sys/signalvar.h>
  55 #include <sys/stat.h>
  56 #include <sys/sx.h>
  57 #include <sys/syscall.h>
  58 #include <sys/sysctl.h>
  59 #include <sys/sysent.h>
  60 #include <sys/vnode.h>
  61
  62 #include <vm/vm.h>
  63 #include <vm/vm_kern.h>
  64 #include <vm/vm_param.h>
  65 #include <vm/pmap.h>
  66 #include <vm/vm_map.h>
  67 #include <vm/vm_object.h>
  68 #include <vm/vm_extern.h>
  69
  70 #include <machine/elf.h>
  71 #include <machine/md_var.h>
  72
  73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
  74 #include <machine/fpu.h>
  75 #include <compat/ia32/ia32_reg.h>
  76 #endif
  77
  78 #define OLD_EI_BRAND    8
  79
  80 static int __elfN(check_header)(const Elf_Ehdr *hdr);
  81 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
  82     const char *interp);
  83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  84     u_long *entry, size_t pagesize);
  85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
  86     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
  87     vm_prot_t prot, size_t pagesize);
  88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
  89
  90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
  91     "");
  92
  93 int __elfN(fallback_brand) = -1;
  94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
  95     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
  96     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
  97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
  98     &__elfN(fallback_brand));
  99
 100 static int elf_trace = 0;
 101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
 102
 103 static int elf_legacy_coredump = 0;
 104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
 105     &elf_legacy_coredump, 0, "");
 106
 107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
 108
 109 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
 110 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
 111 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
 112
 113 int
 114 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
 115 {
 116         int i;
 117
 118         for (i = 0; i < MAX_BRANDS; i++) {
 119                 if (elf_brand_list[i] == NULL) {
 120                         elf_brand_list[i] = entry;
 121                         break;
 122                 }
 123         }
 124         if (i == MAX_BRANDS)
 125                 return (-1);
 126         return (0);
 127 }
 128
 129 int
 130 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
 131 {
 132         int i;
 133
 134         for (i = 0; i < MAX_BRANDS; i++) {
 135                 if (elf_brand_list[i] == entry) {
 136                         elf_brand_list[i] = NULL;
 137                         break;
 138                 }
 139         }
 140         if (i == MAX_BRANDS)
 141                 return (-1);
 142         return (0);
 143 }
 144
 145 int
 146 __elfN(brand_inuse)(Elf_Brandinfo *entry)
 147 {
 148         struct proc *p;
 149         int rval = FALSE;
 150
 151         sx_slock(&allproc_lock);
 152         FOREACH_PROC_IN_SYSTEM(p) {
 153                 if (p->p_sysent == entry->sysvec) {
 154                         rval = TRUE;
 155                         break;
 156                 }
 157         }
 158         sx_sunlock(&allproc_lock);
 159
 160         return (rval);
 161 }
 162
 163 static Elf_Brandinfo *
 164 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
 165 {
 166         Elf_Brandinfo *bi;
 167         int i;
 168
 169         /*
 170          * We support three types of branding -- (1) the ELF EI_OSABI field
 171          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 172          * branding w/in the ELF header, and (3) path of the `interp_path'
 173          * field.  We should also look for an ".note.ABI-tag" ELF section now
 174          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
 175          */
 176
 177         /* If the executable has a brand, search for it in the brand list. */
 178         for (i = 0; i < MAX_BRANDS; i++) {
 179                 bi = elf_brand_list[i];
 180                 if (bi != NULL && hdr->e_machine == bi->machine &&
 181                     (hdr->e_ident[EI_OSABI] == bi->brand ||
 182                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 183                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
 184                         return (bi);
 185         }
 186
 187         /* Lacking a known brand, search for a recognized interpreter. */
 188         if (interp != NULL) {
 189                 for (i = 0; i < MAX_BRANDS; i++) {
 190                         bi = elf_brand_list[i];
 191                         if (bi != NULL && hdr->e_machine == bi->machine &&
 192                             strcmp(interp, bi->interp_path) == 0)
 193                                 return (bi);
 194                 }
 195         }
 196
 197         /* Lacking a recognized interpreter, try the default brand */
 198         for (i = 0; i < MAX_BRANDS; i++) {
 199                 bi = elf_brand_list[i];
 200                 if (bi != NULL && hdr->e_machine == bi->machine &&
 201                     __elfN(fallback_brand) == bi->brand)
 202                         return (bi);
 203         }
 204         return (NULL);
 205 }
 206
 207 static int
 208 __elfN(check_header)(const Elf_Ehdr *hdr)
 209 {
 210         Elf_Brandinfo *bi;
 211         int i;
 212
 213         if (!IS_ELF(*hdr) ||
 214             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
 215             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
 216             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
 217             hdr->e_phentsize != sizeof(Elf_Phdr) ||
 218             hdr->e_version != ELF_TARG_VER)
 219                 return (ENOEXEC);
 220
 221         /*
 222          * Make sure we have at least one brand for this machine.
 223          */
 224
 225         for (i = 0; i < MAX_BRANDS; i++) {
 226                 bi = elf_brand_list[i];
 227                 if (bi != NULL && bi->machine == hdr->e_machine)
 228                         break;
 229         }
 230         if (i == MAX_BRANDS)
 231                 return (ENOEXEC);
 232
 233         return (0);
 234 }
 235
 236 static int
 237 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 238     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
 239 {
 240         struct sf_buf *sf;
 241         int error;
 242         vm_offset_t off;
 243
 244         /*
 245          * Create the page if it doesn't exist yet. Ignore errors.
 246          */
 247         vm_map_lock(map);
 248         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
 249             VM_PROT_ALL, VM_PROT_ALL, 0);
 250         vm_map_unlock(map);
 251
 252         /*
 253          * Find the page from the underlying object.
 254          */
 255         if (object) {
 256                 sf = vm_imgact_map_page(object, offset);
 257                 if (sf == NULL)
 258                         return (KERN_FAILURE);
 259                 off = offset - trunc_page(offset);
 260                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
 261                     end - start);
 262                 vm_imgact_unmap_page(sf);
 263                 if (error) {
 264                         return (KERN_FAILURE);
 265                 }
 266         }
 267
 268         return (KERN_SUCCESS);
 269 }
 270
 271 static int
 272 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 273     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
 274 {
 275         struct sf_buf *sf;
 276         vm_offset_t off;
 277         vm_size_t sz;
 278         int error, rv;
 279
 280         if (start != trunc_page(start)) {
 281                 rv = __elfN(map_partial)(map, object, offset, start,
 282                     round_page(start), prot);
 283                 if (rv)
 284                         return (rv);
 285                 offset += round_page(start) - start;
 286                 start = round_page(start);
 287         }
 288         if (end != round_page(end)) {
 289                 rv = __elfN(map_partial)(map, object, offset +
 290                     trunc_page(end) - start, trunc_page(end), end, prot);
 291                 if (rv)
 292                         return (rv);
 293                 end = trunc_page(end);
 294         }
 295         if (end > start) {
 296                 if (offset & PAGE_MASK) {
 297                         /*
 298                          * The mapping is not page aligned. This means we have
 299                          * to copy the data. Sigh.
 300                          */
 301                         rv = vm_map_find(map, NULL, 0, &start, end - start,
 302                             FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
 303                         if (rv)
 304                                 return (rv);
 305                         if (object == NULL)
 306                                 return (KERN_SUCCESS);
 307                         for (; start < end; start += sz) {
 308                                 sf = vm_imgact_map_page(object, offset);
 309                                 if (sf == NULL)
 310                                         return (KERN_FAILURE);
 311                                 off = offset - trunc_page(offset);
 312                                 sz = end - start;
 313                                 if (sz > PAGE_SIZE - off)
 314                                         sz = PAGE_SIZE - off;
 315                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
 316                                     (caddr_t)start, sz);
 317                                 vm_imgact_unmap_page(sf);
 318                                 if (error) {
 319                                         return (KERN_FAILURE);
 320                                 }
 321                                 offset += sz;
 322                         }
 323                         rv = KERN_SUCCESS;
 324                 } else {
 325                         vm_object_reference(object);
 326                         vm_map_lock(map);
 327                         rv = vm_map_insert(map, object, offset, start, end,
 328                             prot, VM_PROT_ALL, cow);
 329                         vm_map_unlock(map);
 330                         if (rv != KERN_SUCCESS)
 331                                 vm_object_deallocate(object);
 332                 }
 333                 return (rv);
 334         } else {
 335                 return (KERN_SUCCESS);
 336         }
 337 }
 338
 339 static int
 340 __elfN(load_section)(struct vmspace *vmspace,
 341         vm_object_t object, vm_offset_t offset,
 342         caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
 343         size_t pagesize)
 344 {
 345         struct sf_buf *sf;
 346         size_t map_len;
 347         vm_offset_t map_addr;
 348         int error, rv, cow;
 349         size_t copy_len;
 350         vm_offset_t file_addr;
 351
 352         /*
 353          * It's necessary to fail if the filsz + offset taken from the
 354          * header is greater than the actual file pager object's size.
 355          * If we were to allow this, then the vm_map_find() below would
 356          * walk right off the end of the file object and into the ether.
 357          *
 358          * While I'm here, might as well check for something else that
 359          * is invalid: filsz cannot be greater than memsz.
 360          */
 361         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
 362             filsz > memsz) {
 363                 uprintf("elf_load_section: truncated ELF file\n");
 364                 return (ENOEXEC);
 365         }
 366
 367         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
 368         file_addr = trunc_page_ps(offset, pagesize);
 369
 370         /*
 371          * We have two choices.  We can either clear the data in the last page
 372          * of an oversized mapping, or we can start the anon mapping a page
 373          * early and copy the initialized data into that first page.  We
 374          * choose the second..
 375          */
 376         if (memsz > filsz)
 377                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
 378         else
 379                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
 380
 381         if (map_len != 0) {
 382                 /* cow flags: don't dump readonly sections in core */
 383                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
 384                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
 385
 386                 rv = __elfN(map_insert)(&vmspace->vm_map,
 387                                       object,
 388                                       file_addr,        /* file offset */
 389                                       map_addr,         /* virtual start */
 390                                       map_addr + map_len,/* virtual end */
 391                                       prot,
 392                                       cow);
 393                 if (rv != KERN_SUCCESS)
 394                         return (EINVAL);
 395
 396                 /* we can stop now if we've covered it all */
 397                 if (memsz == filsz) {
 398                         return (0);
 399                 }
 400         }
 401
 402
 403         /*
 404          * We have to get the remaining bit of the file into the first part
 405          * of the oversized map segment.  This is normally because the .data
 406          * segment in the file is extended to provide bss.  It's a neat idea
 407          * to try and save a page, but it's a pain in the behind to implement.
 408          */
 409         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
 410         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
 411         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
 412             map_addr;
 413
 414         /* This had damn well better be true! */
 415         if (map_len != 0) {
 416                 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
 417                     map_addr + map_len, VM_PROT_ALL, 0);
 418                 if (rv != KERN_SUCCESS) {
 419                         return (EINVAL);
 420                 }
 421         }
 422
 423         if (copy_len != 0) {
 424                 vm_offset_t off;
 425
 426                 sf = vm_imgact_map_page(object, offset + filsz);
 427                 if (sf == NULL)
 428                         return (EIO);
 429
 430                 /* send the page fragment to user space */
 431                 off = trunc_page_ps(offset + filsz, pagesize) -
 432                     trunc_page(offset + filsz);
 433                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
 434                     (caddr_t)map_addr, copy_len);
 435                 vm_imgact_unmap_page(sf);
 436                 if (error) {
 437                         return (error);
 438                 }
 439         }
 440
 441         /*
 442          * set it to the specified protection.
 443          * XXX had better undo the damage from pasting over the cracks here!
 444          */
 445         vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
 446             round_page(map_addr + map_len),  prot, FALSE);
 447
 448         return (0);
 449 }
 450
 451 /*
 452  * Load the file "file" into memory.  It may be either a shared object
 453  * or an executable.
 454  *
 455  * The "addr" reference parameter is in/out.  On entry, it specifies
 456  * the address where a shared object should be loaded.  If the file is
 457  * an executable, this value is ignored.  On exit, "addr" specifies
 458  * where the file was actually loaded.
 459  *
 460  * The "entry" reference parameter is out only.  On exit, it specifies
 461  * the entry point for the loaded file.
 462  */
 463 static int
 464 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
 465         u_long *entry, size_t pagesize)
 466 {
 467         struct {
 468                 struct nameidata nd;
 469                 struct vattr attr;
 470                 struct image_params image_params;
 471         } *tempdata;
 472         const Elf_Ehdr *hdr = NULL;
 473         const Elf_Phdr *phdr = NULL;
 474         struct nameidata *nd;
 475         struct vmspace *vmspace = p->p_vmspace;
 476         struct vattr *attr;
 477         struct image_params *imgp;
 478         vm_prot_t prot;
 479         u_long rbase;
 480         u_long base_addr = 0;
 481         int vfslocked, error, i, numsegs;
 482
 483         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
 484         nd = &tempdata->nd;
 485         attr = &tempdata->attr;
 486         imgp = &tempdata->image_params;
 487
 488         /*
 489          * Initialize part of the common data
 490          */
 491         imgp->proc = p;
 492         imgp->attr = attr;
 493         imgp->firstpage = NULL;
 494         imgp->image_header = NULL;
 495         imgp->object = NULL;
 496         imgp->execlabel = NULL;
 497
 498         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
 499             curthread);
 500         vfslocked = 0;
 501         if ((error = namei(nd)) != 0) {
 502                 nd->ni_vp = NULL;
 503                 goto fail;
 504         }
 505         vfslocked = NDHASGIANT(nd);
 506         NDFREE(nd, NDF_ONLY_PNBUF);
 507         imgp->vp = nd->ni_vp;
 508
 509         /*
 510          * Check permissions, modes, uid, etc on the file, and "open" it.
 511          */
 512         error = exec_check_permissions(imgp);
 513         if (error)
 514                 goto fail;
 515
 516         error = exec_map_first_page(imgp);
 517         if (error)
 518                 goto fail;
 519
 520         /*
 521          * Also make certain that the interpreter stays the same, so set
 522          * its VV_TEXT flag, too.
 523          */
 524         nd->ni_vp->v_vflag |= VV_TEXT;
 525
 526         imgp->object = nd->ni_vp->v_object;
 527
 528         hdr = (const Elf_Ehdr *)imgp->image_header;
 529         if ((error = __elfN(check_header)(hdr)) != 0)
 530                 goto fail;
 531         if (hdr->e_type == ET_DYN)
 532                 rbase = *addr;
 533         else if (hdr->e_type == ET_EXEC)
 534                 rbase = 0;
 535         else {
 536                 error = ENOEXEC;
 537                 goto fail;
 538         }
 539
 540         /* Only support headers that fit within first page for now      */
 541         /*    (multiplication of two Elf_Half fields will not overflow) */
 542         if ((hdr->e_phoff > PAGE_SIZE) ||
 543             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
 544                 error = ENOEXEC;
 545                 goto fail;
 546         }
 547
 548         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 549         if (!aligned(phdr, Elf_Addr)) {
 550                 error = ENOEXEC;
 551                 goto fail;
 552         }
 553
 554         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 555                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
 556                         prot = 0;
 557                         if (phdr[i].p_flags & PF_X)
 558                                 prot |= VM_PROT_EXECUTE;
 559                         if (phdr[i].p_flags & PF_W)
 560                                 prot |= VM_PROT_WRITE;
 561                         if (phdr[i].p_flags & PF_R)
 562                                 prot |= VM_PROT_READ;
 563
 564                         if ((error = __elfN(load_section)(vmspace,
 565                             imgp->object, phdr[i].p_offset,
 566                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
 567                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
 568                             pagesize)) != 0)
 569                                 goto fail;
 570                         /*
 571                          * Establish the base address if this is the
 572                          * first segment.
 573                          */
 574                         if (numsegs == 0)
 575                                 base_addr = trunc_page(phdr[i].p_vaddr +
 576                                     rbase);
 577                         numsegs++;
 578                 }
 579         }
 580         *addr = base_addr;
 581         *entry = (unsigned long)hdr->e_entry + rbase;
 582
 583 fail:
 584         if (imgp->firstpage)
 585                 exec_unmap_first_page(imgp);
 586
 587         if (nd->ni_vp)
 588                 vput(nd->ni_vp);
 589
 590         VFS_UNLOCK_GIANT(vfslocked);
 591         free(tempdata, M_TEMP);
 592
 593         return (error);
 594 }
 595
 596 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
 597
 598 static int
 599 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
 600 {
 601         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
 602         const Elf_Phdr *phdr, *pnote = NULL;
 603         Elf_Auxargs *elf_auxargs;
 604         struct vmspace *vmspace;
 605         vm_prot_t prot;
 606         u_long text_size = 0, data_size = 0, total_size = 0;
 607         u_long text_addr = 0, data_addr = 0;
 608         u_long seg_size, seg_addr;
 609         u_long addr, entry = 0, proghdr = 0;
 610         int error = 0, i;
 611         const char *interp = NULL, *newinterp = NULL;
 612         Elf_Brandinfo *brand_info;
 613         const Elf_Note *note, *note_end;
 614         char *path;
 615         const char *note_name;
 616         struct sysentvec *sv;
 617
 618         /*
 619          * Do we have a valid ELF header ?
 620          *
 621          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
 622          * if particular brand doesn't support it.
 623          */
 624         if (__elfN(check_header)(hdr) != 0 ||
 625             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
 626                 return (-1);
 627
 628         /*
 629          * From here on down, we return an errno, not -1, as we've
 630          * detected an ELF file.
 631          */
 632
 633         if ((hdr->e_phoff > PAGE_SIZE) ||
 634             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 635                 /* Only support headers in first page for now */
 636                 return (ENOEXEC);
 637         }
 638         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 639         if (!aligned(phdr, Elf_Addr))
 640                 return (ENOEXEC);
 641         for (i = 0; i < hdr->e_phnum; i++) {
 642                 if (phdr[i].p_type == PT_INTERP) {
 643                         /* Path to interpreter */
 644                         if (phdr[i].p_filesz > MAXPATHLEN ||
 645                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
 646                                 return (ENOEXEC);
 647                         interp = imgp->image_header + phdr[i].p_offset;
 648                         break;
 649                 }
 650         }
 651
 652         brand_info = __elfN(get_brandinfo)(hdr, interp);
 653         if (brand_info == NULL) {
 654                 uprintf("ELF binary type \"%u\" not known.\n",
 655                     hdr->e_ident[EI_OSABI]);
 656                 return (ENOEXEC);
 657         }
 658         if (hdr->e_type == ET_DYN &&
 659             (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
 660                 return (ENOEXEC);
 661         sv = brand_info->sysvec;
 662         if (interp != NULL && brand_info->interp_newpath != NULL)
 663                 newinterp = brand_info->interp_newpath;
 664
 665         /*
 666          * Avoid a possible deadlock if the current address space is destroyed
 667          * and that address space maps the locked vnode.  In the common case,
 668          * the locked vnode's v_usecount is decremented but remains greater
 669          * than zero.  Consequently, the vnode lock is not needed by vrele().
 670          * However, in cases where the vnode lock is external, such as nullfs,
 671          * v_usecount may become zero.
 672          */
 673         VOP_UNLOCK(imgp->vp, 0);
 674
 675         error = exec_new_vmspace(imgp, sv);
 676         imgp->proc->p_sysent = sv;
 677
 678         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 679         if (error)
 680                 return (error);
 681
 682         vmspace = imgp->proc->p_vmspace;
 683
 684         for (i = 0; i < hdr->e_phnum; i++) {
 685                 switch (phdr[i].p_type) {
 686                 case PT_LOAD:   /* Loadable segment */
 687                         prot = 0;
 688                         if (phdr[i].p_flags & PF_X)
 689                                 prot |= VM_PROT_EXECUTE;
 690                         if (phdr[i].p_flags & PF_W)
 691                                 prot |= VM_PROT_WRITE;
 692                         if (phdr[i].p_flags & PF_R)
 693                                 prot |= VM_PROT_READ;
 694
 695 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
 696                         /*
 697                          * Some x86 binaries assume read == executable,
 698                          * notably the M3 runtime and therefore cvsup
 699                          */
 700                         if (prot & VM_PROT_READ)
 701                                 prot |= VM_PROT_EXECUTE;
 702 #endif
 703
 704                         if ((error = __elfN(load_section)(vmspace,
 705                             imgp->object, phdr[i].p_offset,
 706                             (caddr_t)(uintptr_t)phdr[i].p_vaddr,
 707                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
 708                             sv->sv_pagesize)) != 0)
 709                                 return (error);
 710
 711                         /*
 712                          * If this segment contains the program headers,
 713                          * remember their virtual address for the AT_PHDR
 714                          * aux entry. Static binaries don't usually include
 715                          * a PT_PHDR entry.
 716                          */
 717                         if (phdr[i].p_offset == 0 &&
 718                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
 719                                 <= phdr[i].p_filesz)
 720                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
 721
 722                         seg_addr = trunc_page(phdr[i].p_vaddr);
 723                         seg_size = round_page(phdr[i].p_memsz +
 724                             phdr[i].p_vaddr - seg_addr);
 725
 726                         /*
 727                          * Is this .text or .data?  We can't use
 728                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
 729                          * alpha terribly and possibly does other bad
 730                          * things so we stick to the old way of figuring
 731                          * it out:  If the segment contains the program
 732                          * entry point, it's a text segment, otherwise it
 733                          * is a data segment.
 734                          *
 735                          * Note that obreak() assumes that data_addr +
 736                          * data_size == end of data load area, and the ELF
 737                          * file format expects segments to be sorted by
 738                          * address.  If multiple data segments exist, the
 739                          * last one will be used.
 740                          */
 741                         if (hdr->e_entry >= phdr[i].p_vaddr &&
 742                             hdr->e_entry < (phdr[i].p_vaddr +
 743                             phdr[i].p_memsz)) {
 744                                 text_size = seg_size;
 745                                 text_addr = seg_addr;
 746                                 entry = (u_long)hdr->e_entry;
 747                         } else {
 748                                 data_size = seg_size;
 749                                 data_addr = seg_addr;
 750                         }
 751                         total_size += seg_size;
 752                         break;
 753                 case PT_PHDR:   /* Program header table info */
 754                         proghdr = phdr[i].p_vaddr;
 755                         break;
 756                 case PT_NOTE:
 757                         pnote = &phdr[i];
 758                         break;
 759                 default:
 760                         break;
 761                 }
 762         }
 763
 764         if (data_addr == 0 && data_size == 0) {
 765                 data_addr = text_addr;
 766                 data_size = text_size;
 767         }
 768
 769         /*
 770          * Check limits.  It should be safe to check the
 771          * limits after loading the segments since we do
 772          * not actually fault in all the segments pages.
 773          */
 774         PROC_LOCK(imgp->proc);
 775         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
 776             text_size > maxtsiz ||
 777             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
 778                 PROC_UNLOCK(imgp->proc);
 779                 return (ENOMEM);
 780         }
 781
 782         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 783         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 784         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 785         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 786
 787         /*
 788          * We load the dynamic linker where a userland call
 789          * to mmap(0, ...) would put it.  The rationale behind this
 790          * calculation is that it leaves room for the heap to grow to
 791          * its maximum allowed size.
 792          */
 793         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
 794             lim_max(imgp->proc, RLIMIT_DATA));
 795         PROC_UNLOCK(imgp->proc);
 796
 797         imgp->entry_addr = entry;
 798
 799         if (interp != NULL) {
 800                 int have_interp = FALSE;
 801                 VOP_UNLOCK(imgp->vp, 0);
 802                 if (brand_info->emul_path != NULL &&
 803                     brand_info->emul_path[0] != '\0') {
 804                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 805                         snprintf(path, MAXPATHLEN, "%s%s",
 806                             brand_info->emul_path, interp);
 807                         error = __elfN(load_file)(imgp->proc, path, &addr,
 808                             &imgp->entry_addr, sv->sv_pagesize);
 809                         free(path, M_TEMP);
 810                         if (error == 0)
 811                                 have_interp = TRUE;
 812                 }
 813                 if (!have_interp && newinterp != NULL) {
 814                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
 815                             &imgp->entry_addr, sv->sv_pagesize);
 816                         have_interp = TRUE;
 817                 }
 818                 if (!have_interp) {
 819                         error = __elfN(load_file)(imgp->proc, interp, &addr,
 820                             &imgp->entry_addr, sv->sv_pagesize);
 821                 }
 822                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 823                 if (error != 0) {
 824                         uprintf("ELF interpreter %s not found\n", interp);
 825                         return (error);
 826                 }
 827         }
 828
 829         /*
 830          * Construct auxargs table (used by the fixup routine)
 831          */
 832         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
 833         elf_auxargs->execfd = -1;
 834         elf_auxargs->phdr = proghdr;
 835         elf_auxargs->phent = hdr->e_phentsize;
 836         elf_auxargs->phnum = hdr->e_phnum;
 837         elf_auxargs->pagesz = PAGE_SIZE;
 838         elf_auxargs->base = addr;
 839         elf_auxargs->flags = 0;
 840         elf_auxargs->entry = entry;
 841         elf_auxargs->trace = elf_trace;
 842
 843         imgp->auxargs = elf_auxargs;
 844         imgp->interpreted = 0;
 845
 846         /*
 847          * Try to fetch the osreldate for FreeBSD binary from the ELF
 848          * OSABI-note. Only the first page of the image is searched,
 849          * the same as for headers.
 850          */
 851         if (pnote != NULL && pnote->p_offset < PAGE_SIZE &&
 852             pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) {
 853                 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
 854                 if (!aligned(note, Elf32_Addr)) {
 855                         free(imgp->auxargs, M_TEMP);
 856                         imgp->auxargs = NULL;
 857                         return (ENOEXEC);
 858                 }
 859                 note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset +
 860                     pnote->p_filesz);
 861                 while (note < note_end) {
 862                         if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) &&
 863                             note->n_descsz == sizeof(int32_t) &&
 864                             note->n_type == 1 /* ABI_NOTETYPE */) {
 865                                 note_name = (const char *)(note + 1);
 866                                 if (strncmp(FREEBSD_ABI_VENDOR, note_name,
 867                                     sizeof(FREEBSD_ABI_VENDOR)) == 0) {
 868                                         imgp->proc->p_osrel = *(const int32_t *)
 869                                             (note_name +
 870                                             round_page_ps(sizeof(FREEBSD_ABI_VENDOR),
 871                                                 sizeof(Elf32_Addr)));
 872                                         break;
 873                                 }
 874                         }
 875                         note = (const Elf_Note *)((const char *)(note + 1) +
 876                             round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) +
 877                             round_page_ps(note->n_descsz, sizeof(Elf32_Addr)));
 878                 }
 879         }
 880
 881         return (error);
 882 }
 883
 884 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
 885
 886 int
 887 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
 888 {
 889         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
 890         Elf_Addr *base;
 891         Elf_Addr *pos;
 892
 893         base = (Elf_Addr *)*stack_base;
 894         pos = base + (imgp->args->argc + imgp->args->envc + 2);
 895
 896         if (args->trace) {
 897                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
 898         }
 899         if (args->execfd != -1) {
 900                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
 901         }
 902         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
 903         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
 904         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
 905         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
 906         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
 907         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
 908         AUXARGS_ENTRY(pos, AT_BASE, args->base);
 909         AUXARGS_ENTRY(pos, AT_NULL, 0);
 910
 911         free(imgp->auxargs, M_TEMP);
 912         imgp->auxargs = NULL;
 913
 914         base--;
 915         suword(base, (long)imgp->args->argc);
 916         *stack_base = (register_t *)base;
 917         return (0);
 918 }
 919
 920 /*
 921  * Code for generating ELF core dumps.
 922  */
 923
 924 typedef void (*segment_callback)(vm_map_entry_t, void *);
 925
 926 /* Closure for cb_put_phdr(). */
 927 struct phdr_closure {
 928         Elf_Phdr *phdr;         /* Program header to fill in */
 929         Elf_Off offset;         /* Offset of segment in core file */
 930 };
 931
 932 /* Closure for cb_size_segment(). */
 933 struct sseg_closure {
 934         int count;              /* Count of writable segments. */
 935         size_t size;            /* Total size of all writable segments. */
 936 };
 937
 938 static void cb_put_phdr(vm_map_entry_t, void *);
 939 static void cb_size_segment(vm_map_entry_t, void *);
 940 static void each_writable_segment(struct thread *, segment_callback, void *);
 941 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
 942     int, void *, size_t);
 943 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
 944 static void __elfN(putnote)(void *, size_t *, const char *, int,
 945     const void *, size_t);
 946
 947 int
 948 __elfN(coredump)(td, vp, limit)
 949         struct thread *td;
 950         struct vnode *vp;
 951         off_t limit;
 952 {
 953         struct ucred *cred = td->td_ucred;
 954         int error = 0;
 955         struct sseg_closure seginfo;
 956         void *hdr;
 957         size_t hdrsize;
 958
 959         /* Size the program segments. */
 960         seginfo.count = 0;
 961         seginfo.size = 0;
 962         each_writable_segment(td, cb_size_segment, &seginfo);
 963
 964         /*
 965          * Calculate the size of the core file header area by making
 966          * a dry run of generating it.  Nothing is written, but the
 967          * size is calculated.
 968          */
 969         hdrsize = 0;
 970         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 971
 972         if (hdrsize + seginfo.size >= limit)
 973                 return (EFAULT);
 974
 975         /*
 976          * Allocate memory for building the header, fill it up,
 977          * and write it out.
 978          */
 979         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 980         if (hdr == NULL) {
 981                 return (EINVAL);
 982         }
 983         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 984
 985         /* Write the contents of all of the writable segments. */
 986         if (error == 0) {
 987                 Elf_Phdr *php;
 988                 off_t offset;
 989                 int i;
 990
 991                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 992                 offset = hdrsize;
 993                 for (i = 0; i < seginfo.count; i++) {
 994                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 995                             (caddr_t)(uintptr_t)php->p_vaddr,
 996                             php->p_filesz, offset, UIO_USERSPACE,
 997                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 998                             curthread);
 999                         if (error != 0)
1000                                 break;
1001                         offset += php->p_filesz;
1002                         php++;
1003                 }
1004         }
1005         free(hdr, M_TEMP);
1006
1007         return (error);
1008 }
1009
1010 /*
1011  * A callback for each_writable_segment() to write out the segment's
1012  * program header entry.
1013  */
1014 static void
1015 cb_put_phdr(entry, closure)
1016         vm_map_entry_t entry;
1017         void *closure;
1018 {
1019         struct phdr_closure *phc = (struct phdr_closure *)closure;
1020         Elf_Phdr *phdr = phc->phdr;
1021
1022         phc->offset = round_page(phc->offset);
1023
1024         phdr->p_type = PT_LOAD;
1025         phdr->p_offset = phc->offset;
1026         phdr->p_vaddr = entry->start;
1027         phdr->p_paddr = 0;
1028         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1029         phdr->p_align = PAGE_SIZE;
1030         phdr->p_flags = 0;
1031         if (entry->protection & VM_PROT_READ)
1032                 phdr->p_flags |= PF_R;
1033         if (entry->protection & VM_PROT_WRITE)
1034                 phdr->p_flags |= PF_W;
1035         if (entry->protection & VM_PROT_EXECUTE)
1036                 phdr->p_flags |= PF_X;
1037
1038         phc->offset += phdr->p_filesz;
1039         phc->phdr++;
1040 }
1041
1042 /*
1043  * A callback for each_writable_segment() to gather information about
1044  * the number of segments and their total size.
1045  */
1046 static void
1047 cb_size_segment(entry, closure)
1048         vm_map_entry_t entry;
1049         void *closure;
1050 {
1051         struct sseg_closure *ssc = (struct sseg_closure *)closure;
1052
1053         ssc->count++;
1054         ssc->size += entry->end - entry->start;
1055 }
1056
1057 /*
1058  * For each writable segment in the process's memory map, call the given
1059  * function with a pointer to the map entry and some arbitrary
1060  * caller-supplied data.
1061  */
1062 static void
1063 each_writable_segment(td, func, closure)
1064         struct thread *td;
1065         segment_callback func;
1066         void *closure;
1067 {
1068         struct proc *p = td->td_proc;
1069         vm_map_t map = &p->p_vmspace->vm_map;
1070         vm_map_entry_t entry;
1071         vm_object_t backing_object, object;
1072         boolean_t ignore_entry;
1073
1074         vm_map_lock_read(map);
1075         for (entry = map->header.next; entry != &map->header;
1076             entry = entry->next) {
1077                 /*
1078                  * Don't dump inaccessible mappings, deal with legacy
1079                  * coredump mode.
1080                  *
1081                  * Note that read-only segments related to the elf binary
1082                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1083                  * need to arbitrarily ignore such segments.
1084                  */
1085                 if (elf_legacy_coredump) {
1086                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1087                                 continue;
1088                 } else {
1089                         if ((entry->protection & VM_PROT_ALL) == 0)
1090                                 continue;
1091                 }
1092
1093                 /*
1094                  * Dont include memory segment in the coredump if
1095                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1096                  * madvise(2).  Do not dump submaps (i.e. parts of the
1097                  * kernel map).
1098                  */
1099                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1100                         continue;
1101
1102                 if ((object = entry->object.vm_object) == NULL)
1103                         continue;
1104
1105                 /* Ignore memory-mapped devices and such things. */
1106                 VM_OBJECT_LOCK(object);
1107                 while ((backing_object = object->backing_object) != NULL) {
1108                         VM_OBJECT_LOCK(backing_object);
1109                         VM_OBJECT_UNLOCK(object);
1110                         object = backing_object;
1111                 }
1112                 ignore_entry = object->type != OBJT_DEFAULT &&
1113                     object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1114                 VM_OBJECT_UNLOCK(object);
1115                 if (ignore_entry)
1116                         continue;
1117
1118                 (*func)(entry, closure);
1119         }
1120         vm_map_unlock_read(map);
1121 }
1122
1123 /*
1124  * Write the core file header to the file, including padding up to
1125  * the page boundary.
1126  */
1127 static int
1128 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1129         struct thread *td;
1130         struct vnode *vp;
1131         struct ucred *cred;
1132         int numsegs;
1133         size_t hdrsize;
1134         void *hdr;
1135 {
1136         size_t off;
1137
1138         /* Fill in the header. */
1139         bzero(hdr, hdrsize);
1140         off = 0;
1141         __elfN(puthdr)(td, hdr, &off, numsegs);
1142
1143         /* Write it to the core file. */
1144         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1145             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1146             td));
1147 }
1148
1149 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1150 typedef struct prstatus32 elf_prstatus_t;
1151 typedef struct prpsinfo32 elf_prpsinfo_t;
1152 typedef struct fpreg32 elf_prfpregset_t;
1153 typedef struct fpreg32 elf_fpregset_t;
1154 typedef struct reg32 elf_gregset_t;
1155 #else
1156 typedef prstatus_t elf_prstatus_t;
1157 typedef prpsinfo_t elf_prpsinfo_t;
1158 typedef prfpregset_t elf_prfpregset_t;
1159 typedef prfpregset_t elf_fpregset_t;
1160 typedef gregset_t elf_gregset_t;
1161 #endif
1162
1163 static void
1164 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1165 {
1166         struct {
1167                 elf_prstatus_t status;
1168                 elf_prfpregset_t fpregset;
1169                 elf_prpsinfo_t psinfo;
1170         } *tempdata;
1171         elf_prstatus_t *status;
1172         elf_prfpregset_t *fpregset;
1173         elf_prpsinfo_t *psinfo;
1174         struct proc *p;
1175         struct thread *thr;
1176         size_t ehoff, noteoff, notesz, phoff;
1177
1178         p = td->td_proc;
1179
1180         ehoff = *off;
1181         *off += sizeof(Elf_Ehdr);
1182
1183         phoff = *off;
1184         *off += (numsegs + 1) * sizeof(Elf_Phdr);
1185
1186         noteoff = *off;
1187         /*
1188          * Don't allocate space for the notes if we're just calculating
1189          * the size of the header. We also don't collect the data.
1190          */
1191         if (dst != NULL) {
1192                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1193                 status = &tempdata->status;
1194                 fpregset = &tempdata->fpregset;
1195                 psinfo = &tempdata->psinfo;
1196         } else {
1197                 tempdata = NULL;
1198                 status = NULL;
1199                 fpregset = NULL;
1200                 psinfo = NULL;
1201         }
1202
1203         if (dst != NULL) {
1204                 psinfo->pr_version = PRPSINFO_VERSION;
1205                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1206                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1207                 /*
1208                  * XXX - We don't fill in the command line arguments properly
1209                  * yet.
1210                  */
1211                 strlcpy(psinfo->pr_psargs, p->p_comm,
1212                     sizeof(psinfo->pr_psargs));
1213         }
1214         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1215             sizeof *psinfo);
1216
1217         /*
1218          * To have the debugger select the right thread (LWP) as the initial
1219          * thread, we dump the state of the thread passed to us in td first.
1220          * This is the thread that causes the core dump and thus likely to
1221          * be the right thread one wants to have selected in the debugger.
1222          */
1223         thr = td;
1224         while (thr != NULL) {
1225                 if (dst != NULL) {
1226                         status->pr_version = PRSTATUS_VERSION;
1227                         status->pr_statussz = sizeof(elf_prstatus_t);
1228                         status->pr_gregsetsz = sizeof(elf_gregset_t);
1229                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1230                         status->pr_osreldate = osreldate;
1231                         status->pr_cursig = p->p_sig;
1232                         status->pr_pid = thr->td_tid;
1233 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1234                         fill_regs32(thr, &status->pr_reg);
1235                         fill_fpregs32(thr, fpregset);
1236 #else
1237                         fill_regs(thr, &status->pr_reg);
1238                         fill_fpregs(thr, fpregset);
1239 #endif
1240                 }
1241                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1242                     sizeof *status);
1243                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1244                     sizeof *fpregset);
1245                 /*
1246                  * Allow for MD specific notes, as well as any MD
1247                  * specific preparations for writing MI notes.
1248                  */
1249                 __elfN(dump_thread)(thr, dst, off);
1250
1251                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1252                     TAILQ_NEXT(thr, td_plist);
1253                 if (thr == td)
1254                         thr = TAILQ_NEXT(thr, td_plist);
1255         }
1256
1257         notesz = *off - noteoff;
1258
1259         if (dst != NULL)
1260                 free(tempdata, M_TEMP);
1261
1262         /* Align up to a page boundary for the program segments. */
1263         *off = round_page(*off);
1264
1265         if (dst != NULL) {
1266                 Elf_Ehdr *ehdr;
1267                 Elf_Phdr *phdr;
1268                 struct phdr_closure phc;
1269
1270                 /*
1271                  * Fill in the ELF header.
1272                  */
1273                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1274                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1275                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1276                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1277                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1278                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1279                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1280                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1281                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1282                 ehdr->e_ident[EI_ABIVERSION] = 0;
1283                 ehdr->e_ident[EI_PAD] = 0;
1284                 ehdr->e_type = ET_CORE;
1285 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1286                 ehdr->e_machine = EM_386;
1287 #else
1288                 ehdr->e_machine = ELF_ARCH;
1289 #endif
1290                 ehdr->e_version = EV_CURRENT;
1291                 ehdr->e_entry = 0;
1292                 ehdr->e_phoff = phoff;
1293                 ehdr->e_flags = 0;
1294                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1295                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1296                 ehdr->e_phnum = numsegs + 1;
1297                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1298                 ehdr->e_shnum = 0;
1299                 ehdr->e_shstrndx = SHN_UNDEF;
1300
1301                 /*
1302                  * Fill in the program header entries.
1303                  */
1304                 phdr = (Elf_Phdr *)((char *)dst + phoff);
1305
1306                 /* The note segement. */
1307                 phdr->p_type = PT_NOTE;
1308                 phdr->p_offset = noteoff;
1309                 phdr->p_vaddr = 0;
1310                 phdr->p_paddr = 0;
1311                 phdr->p_filesz = notesz;
1312                 phdr->p_memsz = 0;
1313                 phdr->p_flags = 0;
1314                 phdr->p_align = 0;
1315                 phdr++;
1316
1317                 /* All the writable segments from the program. */
1318                 phc.phdr = phdr;
1319                 phc.offset = *off;
1320                 each_writable_segment(td, cb_put_phdr, &phc);
1321         }
1322 }
1323
1324 static void
1325 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1326     const void *desc, size_t descsz)
1327 {
1328         Elf_Note note;
1329
1330         note.n_namesz = strlen(name) + 1;
1331         note.n_descsz = descsz;
1332         note.n_type = type;
1333         if (dst != NULL)
1334                 bcopy(&note, (char *)dst + *off, sizeof note);
1335         *off += sizeof note;
1336         if (dst != NULL)
1337                 bcopy(name, (char *)dst + *off, note.n_namesz);
1338         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1339         if (dst != NULL)
1340                 bcopy(desc, (char *)dst + *off, note.n_descsz);
1341         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1342 }
1343
1344 /*
1345  * Tell kern_execve.c about it, with a little help from the linker.
1346  */
1347 static struct execsw __elfN(execsw) = {
1348         __CONCAT(exec_, __elfN(imgact)),
1349         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1350 };
1351 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));